AI Safety

What AI providers state their models reached in safety testing, in their own words, and what outside testers found.

Also: compare models side by side · statements that a capability level was crossed · providers · practices providers document · safety news.

What providers state

Each row is one statement in a provider's own safety document about a capability level of one of its models: the level as the provider names it, whether the provider says the model reached it, and the sentence that says so.

138 of 138

Model	Level, as the provider names it	The provider states	The provider's sentence
GPT-5.5 OpenAI · 2026-04-23	High capability, below critical Cyber	Not reached	“we are treating GPT-5.5 as High capability in the Cybersecurity domain, but below Critical.” source
Claude Opus 4.7 Anthropic · 2026-04-16	Automated AI R&D Autonomy and AI research	Not reached	“that Opus 4.7 does not cross the threshold for automated AI R&D;” source
Claude Opus 4.7 Anthropic · 2026-04-16	CB-1 Chemical, biological, radiological, nuclear	Not determined	“We believe it is hard to be conﬁdent regarding whether a model passes this threshold.” source
Claude Opus 4.7 Anthropic · 2026-04-16	CB-2 Chemical, biological, radiological, nuclear	Not determined	“Our evaluations suggest that the model is not yet at the level of capability associated with the CB-2 threat model (above).” source
Claude Mythos Preview Anthropic · 2026-04-07	AI R&D rule-out Autonomy and AI research	Not reached	“The main reason we have determined that Claude Mythos Preview does not cross the threshold in question is that we have been using it extensively in the course of our day-to-day work and exploring where it can automate such work, and it does not seem close to being able to substitute for Research Scientists and Research Engineers—especially relatively senior ones.” source
Claude Mythos Preview Anthropic · 2026-04-07	Automated AI R&D Autonomy and AI research	Not reached	“We assess that Claude Mythos Preview does not cross the automated AI-R&D capability threshold.” source
Claude Mythos Preview Anthropic · 2026-04-07	RSP automated AI R&D Autonomy and AI research	Not reached	“we have concluded that Claude Mythos Preview does not cross the RSP automated AI R&D threshold of compressing two years of progress into one.” source
Claude Mythos Preview Anthropic · 2026-04-07	CB-1 Chemical, biological, radiological, nuclear	Not determined	“We believe it is hard to be conﬁdent regarding whether a model passes this threshold.” source
Claude Mythos Preview Anthropic · 2026-04-07	CB-2 Chemical, biological, radiological, nuclear	Not reached	“We believe that Claude Mythos Preview does not pass this threshold due to its noted limitations in open-ended scientiﬁc reasoning, strategic judgment, and hypothesis triage.” source
GPT-5.4 mini OpenAI · 2026-03-05	High Chemical, biological, radiological, nuclear	Not reached	“GPT-5.4 mini’s biology capabilities are closely comparable to those of GPT-5. We conclude that it, too, is not High under our Preparedness Framework.” source
GPT-5.4 Thinking OpenAI · 2026-03-05	High Autonomy and AI research	Not reached	“For AI self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.4-thinking does not have a plausible chance of reaching a High threshold.” source
GPT-5.4 Thinking OpenAI · 2026-03-05	High Chemical, biological, radiological, nuclear	Reached	“As we did for gpt-5.2-thinking before it, we are continuing to treat gpt-5.4-thinking as High capability in the Biological and Chemical domain.” source
GPT-5.4 Thinking OpenAI · 2026-03-05	High Cyber	Reached	“As we did for gpt-5.3-codex, we are treating gpt-5.4-thinking as High capability in the Cybersecurity domain, and applied safeguards as described in the Safeguard section below.” source
Claude Sonnet 4.6 Anthropic · 2026-02-17	AI R&D-4 Autonomy and AI research	Not reached	“Our determination is that Claude Sonnet 4.6 does not cross either the AI R&D-4 or the CBRN-4 capability threshold.” source
Claude Sonnet 4.6 Anthropic · 2026-02-17	ASL-3 Chemical, biological, radiological, nuclear	Not determined	“Claude Sonnet 4.6 performed above thresholds for ASL-3 rule-in on all three evaluations for ASL-3, and it did not exceed the performance of previous models.” source
Claude Sonnet 4.6 Anthropic · 2026-02-17	ASL-4 Chemical, biological, radiological, nuclear	Not reached	“In particular, it did not cross the threshold on our ASL-4 rule-out evaluation short-horizon computational biology tasks evaluation.” source
Claude Sonnet 4.6 Anthropic · 2026-02-17	CBRN-4 Chemical, biological, radiological, nuclear	Not reached	“Our determination is that Claude Sonnet 4.6 does not cross either the AI R&D-4 or the CBRN-4 capability threshold.” source
GPT-5.3-Codex OpenAI · 2026-02-05	High Autonomy and AI research	Not reached	“It does not reach High capability on AI self-improvement.” source
GPT-5.3-Codex OpenAI · 2026-02-05	High Chemical, biological, radiological, nuclear	Reached	“We are also treating GPT-5.3-Codex as High risk in the Biological and Chemical domain, and applying the same safeguards as for our other deployments of models that reach this designation.” source
GPT-5.3-Codex OpenAI · 2026-02-05	High Cyber	Not determined	“This is the first launch we are treating as High capability in the Cybersecurity domain under our Preparedness Framework, activating the associated safeguards.” source
Gemini 3 Pro Google DeepMind · 2025-11	Uplift level 1 Cyber	Not determined	“Cybersecurity On key skills benchmark, v1 hard challenges: 11/12 challenges solved; v2 challenges: 0/13 solved end-to-end.” source
Gemini 3 Pro Google DeepMind · 2025-11	Instrumental reasoning level 1 + 2 Deception and sabotage	Not determined	“Misalignment (Exploratory) Agent solves 3/11 situational awareness challenges and 1/4 stealth challenges.” source
Gemini 3 Pro Google DeepMind · 2025-11	Level 1 (exploratory) Deception and sabotage	Not reached	“Harmful Manipulation Model manipulative eﬃcacy improves on non-generative AI baseline, but shows no signiﬁcant uplift versus prior models and does not reach alert thresholds.” source
o4-mini OpenAI · 2025-04-16	High Autonomy and AI research	Not reached	“openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.” source
o4-mini OpenAI · 2025-04-16	High Chemical, biological, radiological, nuclear	Not reached	“openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.” source
o4-mini OpenAI · 2025-04-16	High Cyber	Not reached	“openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.” source
o3 OpenAI · 2025-04-16	High Autonomy and AI research	Not reached	“openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.” source
o3 OpenAI · 2025-04-16	High Chemical, biological, radiological, nuclear	Not reached	“openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.” source
o3 OpenAI · 2025-04-16	High Cyber	Not reached	“openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.” source
Operator OpenAI · 2025-01-23	Low Autonomy and AI research	Not reached	“we adapted these two evals for the computer-use setting and found the pre-mitigation operator model to be “low” risk in both categories, consistent with the gpt‑4o model.” source
Operator OpenAI · 2025-01-23	Low Chemical, biological, radiological, nuclear	Not reached	“this 1% qualifies as “low” risk, matching gpt‑4o’s “low” risk level, as detailed in openai’s preparedness framework (opens in a new window) 3 .” source
Muse Spark Contemplating Meta	Moderate or lower Autonomy and AI research	Not reached	“Loss of Control risks.We assess loss of control risk as “moderate or lower.”” source
Muse Spark Contemplating Meta	High Chemical, biological, radiological, nuclear	Reached	“We retain the same “high risk” determination for the unmitigated model.” source
Muse Spark Contemplating Meta	Moderate or lower Chemical, biological, radiological, nuclear	Not reached	“After validating that the mitigations deployed for Muse Spark, including refusal mechanisms, continue to perform robustly under the Contemplating architecture, we confirm residual risk at “moderate or lower.”” source
Muse Spark Contemplating Meta	Moderate or lower Cyber	Not reached	“We assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in Meta’s Advanced AI Scaling Framework.” source
Muse Spark 1.1 Meta	Capability checkpoint Autonomy and AI research	Reached	“Muse Spark 1.1 demonstrates improved coding capabilities, which are now meeting one of the indicative thresholds introduced in Muse Spark Safety & Preparedness Report (specifically, resolving over half of unique tasks on SWE-Bench Verified Hard).” source
Muse Spark 1.1 Meta	Moderate or lower Autonomy and AI research	Not reached	“In the Loss of Control domain, the model’s capabilities remain within the “moderate or lower” threshold under Meta’s Advanced AI Scaling Framework.” source
Muse Spark 1.1 Meta	High Chemical, biological, radiological, nuclear	Not determined	“Our assessments indicate that, when evaluated without mitigations applied, we cannot rule out Muse Spark 1.1’s capabilities meeting this threshold in both the Chemical & Biological and Cybersecurity domains.” source
Muse Spark 1.1 Meta	High Cyber	Not determined	“Our assessments indicate that, when evaluated without mitigations applied, we cannot rule out Muse Spark 1.1’s capabilities meeting this threshold in both the Chemical & Biological and Cybersecurity domains.” source
Muse Spark 1.1 Meta	Moderate or lower Cyber	Not reached	“Therefore, we conclude that we have addressed these risks from unmitigated Muse Spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under Advanced AI Scaling Framework.” source

Previous · Page 2 of 4 · Next

What outside testers report

Each row is one result from a government evaluator, independent lab, academic or security firm that tested a model, quoted from where it is reported.

1911 of 1911

Model	Tester	Test	What was found
Grok 4.6 Cursor	Independent lab Independent lab · 2026-08-12	vals index Other	“Vals Index (Vals AI) is an independent composite of real-world industry and agentic evaluations spanning finance, legal, healthcare, and coding-adjacent professional work.” source
Grok 4.6 Cursor	Independent lab Independent lab · 2026-08-12	harvey legal agent benchmark Other	“Harvey Legal Agent Benchmark (LAB) is a long-horizon legal-agent evaluation: agents complete realistic, multi-step legal work over client-matter files and produce reviewable work product graded with expert all-pass rubrics.” source
Grok 4.6 Cursor	Independent lab Independent lab · 2026-08-12	third-party evaluator testing of cyber capabilities Cyber	“We additionally supplied the model without restrictions to third-party evaluators to validate the above results.” source
Solar Pro 4 Upstage	Independent lab Independent lab · 2026-08-11	τ³-banking Prompt injection and agents	“Multi-turn tool use (τ³-Banking): 23” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	LLM agent-BT evaluation Chemical, biological, radiological, nuclear	“In this report, we evaluated LLM Agents on their use of BTs to redesign biomolecules to evade nucleic acid synthesis screening measures.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	nucleic acid synthesis screening evasion testing of agent-designed DNA sequences Chemical, biological, radiological, nuclear	“The DNA sequences designed by our Agents were then tested to gauge whether they would be able to evade nucleic acid synthesis screening measures.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	biomolecule redesign task Chemical, biological, radiological, nuclear	“In three out of 30 instances, all of which were CYOA runs, Gemini’s redesigned Protein-2 was able to pass the entirety of our scoring pipeline, including evasion of a commercial sequence screening process (Figure 2.2).” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	simulated screening mockup Chemical, biological, radiological, nuclear	“Interestingly, although Gemini’s redesigned OmpA proteins passed validation and structural scoring gates more frequently than any other biomolecule (20/30 attempts), they met with a nearly 100% failure rate at the screening gate, with only one OmpA redesign passing our simulated screening check.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	biomolecule redesign evaluation Chemical, biological, radiological, nuclear	“Similar to Gemini’s results, we found that three of 30 redesign attempts resulted in Protein-2 redesigns that passed all grading checkpoints and were able to bypass commercial nucleic acid synthesis screening (Figure 2.4).” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	peptide-2 redesign Chemical, biological, radiological, nuclear	“For both peptides, DeepSeek’s performance characteristics were similar to those of Gemini: it performed relatively well on glucagon redesign (14/30 total attempts were successful) and succeeded in a handful of attempts with Peptide-2 in the ProteinMPNN environment (Figures 2.3 and 2.4).” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	ompA protein redesign Chemical, biological, radiological, nuclear	“Though both models had similarly high overall rates of OmpA redesign success on checks for validation and for structure and function, DeepSeek’s OmpA protein redesign was significantly more successful at evading our simulated screening pipeline (Figure 2.4).” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	viral genome redesign Chemical, biological, radiological, nuclear	“We tested DeepSeek V4 Pro on this task in the CYOA BT configuration.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	protein-2 redesign for screening evasion Chemical, biological, radiological, nuclear	“Gemini 3.1 Pro was able to successfully redesign Protein-2 for screening evasion in three of 30 attempts.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	simulated screening process Chemical, biological, radiological, nuclear	“For all but one of our test biomolecules, we implemented a simulated screening process that utilized the NCBI BLAST tool.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	Parvovirus B19 redesign evaluation Chemical, biological, radiological, nuclear	“We did not implement the function check component of our full viral genome here, as all Agent submissions failed the preceding validity or structure checks.” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	viral genome redesign and screening evasion case study Chemical, biological, radiological, nuclear	“Appendix E: Viral genome redesign and screening evasion case study” source
Gemini 3.1 Pro Google DeepMind	RAND (CAST / Project Canary) Independent lab · 2026-08-11	per-criterion scoring results Chemical, biological, radiological, nuclear	“The heatmaps of Figures D.1 and D.2 show the number of attempts, out of 10, in which Gemini 3.1 Pro and DeepSeek V4 Pro achieved each of our validity, structure, and function- related scoring criteria.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	peptide and protein redesign task Chemical, biological, radiological, nuclear	“We tested autonomous Agents, driven by frontier LLMs, on their ability to use BTs for peptide and protein redesign tasks that could potentially enable evasion of nucleic acid screening measures.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	nucleic acid screening pipelines Chemical, biological, radiological, nuclear	“Sequences that scored highly on those metrics were evaluated by nucleic acid screening pipelines to determine whether the Agent had sufficiently obfuscated the sequence that encoded the desired biomolecule so that it could evade the screening measures in place at some major commercial nucleic acid synthesis providers.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	LLM agent-BT evaluation (DNA sequence redesign task to evade nucleic acid synthesis screening) Chemical, biological, radiological, nuclear	“In this report, we evaluated LLM Agents on their use of BTs to redesign biomolecules to evade nucleic acid synthesis screening measures.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	biomolecule redesign task (peptide and protein redesign to evade sequence screening) Chemical, biological, radiological, nuclear	“We found that both Gemini 3.1. Pro and DeepSeek V4 Pro were able to redesign peptides or proteins to evade sequence screening in several attempts.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	nucleic acid synthesis screening (simulated screening mockup and commercial screening algorithm) Chemical, biological, radiological, nuclear	“The peptide and protein redesigns generated by our Agents were screened using one of two pipelines: three sets of Agent designs were passed through a simulated screening mockup, while Protein-2 designs were instead tested with a commercial screening algorithm.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	biomolecule redesign evaluation (protein-2, peptide-2, glucagon, OmpA redesign across CYOA, ProteinMPNN, Evo 2, and RFdiffusion + ProteinMPNN configurations) Chemical, biological, radiological, nuclear	“Similar to Gemini’s results, we found that three of 30 redesign attempts resulted in Protein-2 redesigns that passed all grading checkpoints and were able to bypass commercial nucleic acid synthesis screening (Figure 2.4).” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	glucagon redesign Chemical, biological, radiological, nuclear	“For both peptides, DeepSeek’s performance characteristics were similar to those of Gemini: it performed relatively well on glucagon redesign (14/30 total attempts were successful) and succeeded in a handful of attempts with Peptide-2 in the ProteinMPNN environment (Figures 2.3 and 2.4).” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	viral genome redesign (parvovirus B19 proteins) Chemical, biological, radiological, nuclear	“We tested DeepSeek V4 Pro on this task in the CYOA BT configuration.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	parvovirus B19 full viral genome redesign task Chemical, biological, radiological, nuclear	“We found that the DeepSeek V4 Pro Agent could not successfully redesign all parvovirus B19 proteins.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	simulated screening process for redesigned viral proteins Chemical, biological, radiological, nuclear	“Four out of ten redesigns were able to evade screening across all proteins (Figure E.2).” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	protein-2 redesign task Chemical, biological, radiological, nuclear	“DeepSeek V4 Pro also succeeded at the Protein-2 redesign task in three of 30 attempts.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	nucleic acid synthesis screening test Chemical, biological, radiological, nuclear	“In this report, we made use of a commercial nucleic acid synthesis vendor’s screening system and our own simulated screening pipeline, but a number of other screening approaches exist.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	synthesis screening evasion tests Chemical, biological, radiological, nuclear	“In our synthesis screening evasion tests, we notified our research partner of our intention and research objectives, then worked with them to confirm sequence submission details.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	biomolecule redesign tasks Chemical, biological, radiological, nuclear	“We paired each of our biomolecule redesign tasks with BT environments as described in Table A.1.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	sequence screening checkpoint Chemical, biological, radiological, nuclear	“The DNA sequences designed by the Agents were subject to the sequence screening checkpoint described in Appendix C.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	validity checks Chemical, biological, radiological, nuclear	“Validity checks were used to determine whether the generated DNA sequences were biologically coherent and useful for the task.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	structure and function metric thresholds Chemical, biological, radiological, nuclear	“Structure and function metric thresholds were set to determine whether the biomolecules encoded by the generated DNA sequences retained general folding structures similar to those of their wildtype templates.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	simulated screening process that utilized the ncbi blast tool Chemical, biological, radiological, nuclear	“For all but one of our test biomolecules, we implemented a simulated screening process that utilized the NCBI BLAST tool.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	commercial screening pipeline Chemical, biological, radiological, nuclear	“For Protein-2, which would be expected to be identified by commercial screening processes, we worked with a nucleic acid synthesis provider to utilize their screening pipeline for testing redesigns.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	viral genome redesign and screening evasion case study Chemical, biological, radiological, nuclear	“We did not implement the function check component of our full viral genome here, as all Agent submissions failed the preceding validity or structure checks.” source
DeepSeek-V4-Pro DeepSeek	RAND (CAST / Project Canary) Independent lab · 2026-08-11	validity, structure, and function-related scoring criteria Chemical, biological, radiological, nuclear	“The heatmaps of Figures D.1 and D.2 show the number of attempts, out of 10, in which Gemini 3.1 Pro and DeepSeek V4 Pro achieved each of our validity, structure, and function- related scoring criteria.” source
Ling-3.0-tiny InclusionAI (Ant Group)	Independent lab Independent lab · 2026-08-10	Artificial Analysis Intelligence Index v4.1.1 Other	“Ling-3.0-tiny achieves a score of 25 on the Artificial Analysis Intelligence Index v4.1.1 and 16 on the Artificial Analysis Agentic Index.” source
Ling-3.0-tiny InclusionAI (Ant Group)	Independent lab Independent lab · 2026-08-10	Terminal-Bench 2.1 Other	“Terminal-Bench 2.1: Evaluated under the Artificial Analysis (AA) protocol using the default Terminus 2 harness, a unified 2-hour timeout, the provided JSON parser in preserve-thinking mode, and 3 runs per task (mean).” source

Previous · Page 13 of 48 · Next